RΛIN / ATO · Research in public · 06 November 2026
Research frontiers.
Claims meet tests.
Seven public investigations. A prediction, a visible result and an honest limit for each.
The Observatory connects ideas through questions that can be tested. Follow the experiment, record what changes and keep the assumptions beside the result.
The public research program
Choose a question.
Keep the evidence.
01 / Language
Cipher boundaries
Reveal known unit boundaries and count compatible parses. The synthetic generating key makes the answer checkable.
Limit: A synthetic success is not a Voynich decipherment.
Run the investigation →02 / Matter
Mechanical memory
Hold the stored word fixed and perturb its reader. Test whether recalibration recovers the original reading.
Limit: An affine sensor demonstration is not a physical device validation.
Run the investigation →03 / Chemistry
Seed capability
Compare reachability and deadline hits from different seeds in an assumed reaction network.
Limit: Simulation under stated rules does not establish autonomous life.
Run the investigation →04 / Algorithms
Selective classification
Measure coverage and accepted accuracy together under a synthetic distribution shift.
Limit: Rejecting more cases does not guarantee reliable accepted answers.
Run the investigation →05 / Geometry
Symbol geometry
Move a four-coordinate state while its Hopf base stays fixed. Compare all coordinates.
Limit: A shared image does not identify a unique higher-dimensional source.
Run the investigation →06 / Mathematics
Golden Scale
Scale a positive vector and iterate the Fibonacci transformation. Compare its magnitude with the retained ratio.
Limit: The recurrence is classical; a physical golden law needs separate evidence.
Run the investigation →07 / Information
Context repair
Find same-summary, different-answer pairs in an eight-state model. Test which extra bits remove the ambiguity.
Limit: Sufficiency depends on the declared state set and question.
Run the investigation →Five gates before a broader claim
Make the next test
discriminate.
- Define State the system, observable and claim precisely enough to be wrong.
- Separate Identify competing explanations and a measurement on which they disagree.
- Control Compare against a baseline, a negative control and the declared assumptions.
- Replicate Test on genuinely unseen cases; keep training, tuning and evaluation distinct.
- Preserve Publish methods, uncertainty, failed branches and versioned observations.
Geometry / interpretation
What would reveal
the hidden state?
If every allowed observable factors through one projection, states with the same projection remain indistinguishable through those observables. A two-dimensional yin–yang image by itself cannot identify a unique four-dimensional source or certify wormhole topology.
The useful next step is to name a candidate model and a separating observable. Symbolic resemblance can inspire that question; it cannot supply the measurement.
Inspect the yin–yang candidate → · Explore genuine 4D geometry →
Your investigation desk
A trail you can revisit.
Record a prediction before opening a lab. Capture more than one setting, compare the observations, and save your interpretation. The notebook stays on your device and can be exported as JSON.
Begin an investigation → · Browse 26 connected questions →
The two new model sources are public: Golden Scale and Context Repair source. Classical background is cited on each experiment page.